search

autonomousvision / unimatch

[TPAMI'23] Unifying Flow, Stereo and Depth Estimation
https://haofeixu.github.io/unimatch/
MIT License
1.02k stars 105 forks source link

Training for depth on my own custom dataset #62

Closed YacineDeghaies closed 1 month ago

YacineDeghaies commented 1 month ago

Hi, I'd like to train the model to estimate depth using my own dataset that is made up of only RGB images and their ground truth depth maps. Can you please give me any hints on how to tackle this ? Thank you

I thought modifying the DemonDataset class..

haofeixu commented 1 month ago

Hi, you will need to write the code to load the images, intrinsic & extrinsic matrices, and the corresponding grouth truth. Yes you can modify the __init__ func in the DemonDataset class.

YacineDeghaies commented 1 month ago

Thanks ! For the camera intrinsic and extrinsic parameters, do I need to use COLMAP just like you mentioned in the paper ? because I don't have them.

my images have 1080x1920 as a resolution, would that not be a problem?

haofeixu commented 1 month ago

Yes, we assume intrinsic and extrinsic parameters are known. However, if your image resolution is 1080x1920, our model so far is not able to handle such high-resolution images very efficiently.

YacineDeghaies commented 1 month ago

Thank you for helping !